Purge Baby Purge, Jordan Wrong on China AI, Osnos Wrong on ‘China Malaise,’ Brain-Inspired AI Algos Won’t Work
+ My Favorite Rabbis on October 7
Before we get into the China and tech content today, I wanted to highlight a podcast I put out about the October 7 attacks in Israel.
It’s been difficult to express just how shattering they have been to people in my life who aren’t Jewish. This episode, where I host two of America’s leading young rabbis, Zohar Atkins and Ari Lamm, is my attempt at trying to provide some solace to fellow Jews and convey its impact to a broader audience. I’m a little scared to share this with you all, but I hope you find some time to listen with openness and generosity.
Why Jordan Is Wrong About China’s AI Prospects
By anonymous columnist L-Squared, who recently wrote a piece for ChinaTalk on Hugging Face’s end in the China market.
In a recent essay, Jordan and Matthew Mittelsteadt argue that AI diffusion will be key to national competitiveness for the coming decades. While I’m sympathetic to that argument, there were a few parts of the piece that missed the mark. With thanks to Jordan for giving me the opportunity to take him down on his own Substack, let’s dig into each of these in turn!
First, the essay claims that “top AI researchers can be enticed from across the Pacific by top Chinese firms offering salaries in the millions, or even to Chinese firms’ Bay Area research labs.” In reality, talent flows are clearly not in China’s favor. An analysis of authors accepted to a major machine-learning conference in 2019 found that a majority of top-tier Chinese researchers had left China for the US. Recent evidence suggests tech professionals constitute a disproportionate share of the rising numbers of people leaving China. (Who knows, if my smartest friends weren’t departing Beijing in droves, maybe I’d be out having fun instead of writing takedowns like this!) Meanwhile, work-culture challenges seem to limit the ability of Chinese companies’ Bay Area labs to attract and retain diverse talent.
Second, the authors write that “Beijing is embracing an unexpectedly light regulatory touch” to large language models. Well sure — if you ignore the fact that the public-facing deployment of LLMs requires regulator sign-off and the first batch of approvals involved frequent meetings with regulators and constant document submissions. While the process may become more streamlined, on current evidence China’s approach is far from light-touch. Describing it as such risks encouraging overly permissive AI regulation in the US.
Finally, framing AI development as a “race” — as the essay’s headline does — is wrong and dangerous. Justin Sherman outlines a couple of reasons, to which I’ll add: if the authors are correct that technological diffusion matters more than innovating first, why perpetuate a metaphor that emphasizes speed over spread?
What Osnos’s ‘Age of Malaise’ Piece Misses about China Today
Steven Zhao is a writer studying economics and computer science at Stanford.
There are two main problems with Evan Osnos’s recent New Yorker feature about the mood in China today. Mainly, it extrapolates the views of a few elites and people in Beijing to the entirety of China, conflating their opinions with the broader urban middle class.
For instance, take lying flat 躺平. The phenomenon began during the frenzied zenith of China’s tech sector, not its crackdown. It was a response not to CCP oppression, but rather exploitative bosses — Osnos’s very interviewees in this piece. Recall that Jack Ma himself told young folks that 996 was good. As such, the middle-class youth’s economic frustration stems not from suppression of the private sector, but rather its grotesque power. While the economic slowdown has certainly exacerbated this frustration, such discontent is rooted in the frenetic go-go days.
It doesn’t make sense to treat the urban middle class and the business elite as the same group, because their interests are often diametrically opposed. The former was struggling from the machinations of the old system, while the latter thrived. Yet foreign commentators seem to continuously lump these two groups together. Perhaps one reason is that lots of young urban folks are cosmopolitan and educated — as are many businesspeople — so many people assume the reasons for their opposition against the CCP are the same. This assumption, however, usually doesn’t hold true — except for young urban people who are also very rich — and we should be more careful about parsing the nuances.
Can China Beat Export Controls With Creative Chip Design?
James Wang is a partner at Creative Ventures.
[Jordan: export controls are changing the fitness landscape, pushing down the peak of “normal” AI chips which make other options more competitive and worthy of time and money to explore.]
The US is forcing China into a corner where, in coming years, Chinese engineers will be pushed to explore different technological paradigms. Biologically based or non-parallel structures for training and inference, however, are probably a dead end.
The idea of building AI systems that look like the brain isn’t new. It’s what drove the original concept of perceptrons and perceptron networks, and it’s what eventually became modern neural networks. Researchers also tried to represent models in semantic form in knowledge-based AI systems and symbolic systems. We did this in the ‘70s, ‘80s, and ‘90s — all research pathways which led to various AI winters.
One of the main triumphs through the 2000s and especially into the 2010s was not only a growth in computing power, but a philosophical shift in the field to “do what works” as opposed to limiting research to more conceptually neat systems. Over the years, we moved away from activation/no-activation (how our neurons work) to continuous functions. Today, researchers use backpropagation for training neural networks, a mathematical convenience that has no biological analog. And of course, transformers are a brilliant but entirely contrived concept to make it easier to parallelize auto-regressive models.
What’s more: parallelism is a useful structure far beyond just GPUs. It runs all the way down to GPUs and machine-level parallel instructions, as well as all the way up to multiple machines, racks, and data centers (more on this here). Parallelism is pretty much here to stay.
I’ve seen a lot of low-probability but imaginative work on new types of chip fabbing processes, optical computing, and so on. I’d expect that being forced out of our current hardware stack is more likely to push China to explore those areas, especially since a lot of their strong talent is already in the hardware/semiconductor space, as SMIC/Huawei have demonstrated.
James writes the following substack:
Purge Baby Purge! The Politics Behind China’s Disappearing Leaders
Finally, a guest post by Mark Parker Young, a former CIA analyst now at Mandiant and a non-resident fellow at the Atlantic Council. This piece first appeared on the Atlantic Council’s website. These views don’t reflect those of Mark’s employers.
Chinese Communist Party (CCP) General Secretary Xi Jinping has shaken China’s military and foreign affairs establishments in the past two months by abruptly replacing several senior military officers and China’s minister of foreign affairs. The removals were all the more surprising because Xi had promoted many of these same officials to lead their organizations less than a year earlier. A close look at the officials involved suggests that a variety of personal and institutional factors contributed to their downfall, but the disruptive impact of the sudden disappearances indicates underlying mistakes and misjudgments on the part of Xi and the personnel apparatus he oversees.
The recent removals suggest that Xi has approved prosecutions of several discrete pockets of corruption and misconduct rather than a repeat of the sweeping and interconnected purges of his first term. The senior officials involved had crucial roles within their respective military and civilian bureaucracies, but none was part of Xi’s core apparatus of political control.
Interpreting patterns among ousted officials
The reshuffles in the People’s Liberation Army (PLA) constitute its most significant internal upheaval since 2017. Recent anti-corruption investigations appear to be radiating outward from the traditional locus of military corruption: procurement and logistics. In the last two months, investigators have reportedly detained Minister of National Defense Li Shangfu, Rocket Force Commander Li Yuchao, Rocket Force Political Commissar Xu Zhongbo, and several of their deputies. Li Shangfu served from 2017 to 2022 as chief of the PLA’s armaments and procurement department and the Rocket Force is an extremely capital-intensive service that has expanded rapidly in the past decade, likely affording numerous opportunities for graft. Xu also previously served as political commissar of the Joint Logistics Department and is the latest in a long line of its former leaders to fall under suspicion. The new Rocket Force leaders have no prior experience with the force and its incoming political commissar significantly outranks the new commander in the CCP hierarchy, signaling Xi’s determination to uproot their predecessors’ personal networks and reimpose discipline.
By targeting procurement-related organs, Xi has launched a fresh campaign against one of the two military institutions most susceptible to corruption; the other is the political work system and the network of political commissars embedded at every level of the PLA. If prosecutions were to expand beyond procurement-linked officers to implicate broader networks within the Central Military Commission (CMC) Political Work Department, then the disruptive impact would likely spread across the PLA.
The circumstances of Foreign Minister Qin Gang’s removal remain unclear, but they are more likely to stem from personal misconduct and idiosyncratic factors that have a narrower impact on the national security establishment. Qin’s own network within the Ministry of Foreign Affairs (MFA) may be somewhat narrow because of his unusual and rapid ascent. MFA officials also have limited influence over major contracts and assets, so presumably less direct opportunity for large-scale graft. If Xi decides to publicly charge Qin with a broad range of offenses, however, investigators may find grist in his prior service in the MFA Protocol Department, where he would have been responsible for the disposition of official gifts, travel, and hosting functions. One of the few senior Chinese diplomats charged with corruption in recent years was former Assistant Minister of Foreign Affairs Zhang Kunsheng, who also led protocol functions and was dismissed from his position in 2015.
Alongside these high-profile removals, the internal investigations apparatus continues to churn through the middle ranks of the civilian sector under the new secretary of the Central Commission for Discipline Inspection, Li Xi. Some of the targets this year have been significant, such as the former Guizhou party secretary, former secretary of the Bank of China’s party committee, and a raft of mid-level officials from the discipline inspection system itself. However, there is no clear thread that links these investigations to top leaders.
Removals point to vetting missteps and misjudgments
Xi’s management of the CCP’s personnel system is vital to his political position and to the functioning of the party-state. At the same time, the identity of the recently removed officials along with their brief tenures suggest that he has committed unforced errors.
High-level prosecutions are not in and of themselves a sign of misjudgment, because there is a powerful rationale for Xi to periodically prosecute officials in politically sensitive positions. Working as a senior official in the Ministry of Public Security’s headquarters, for example, is likely to remain a high-risk occupation as long as Xi is in office. Xi has an incentive to selectively undermine ties and trust between officials who are in a position to affect his political security. Indeed, he has already demonstrated his willingness to use members of the security establishment against one another. Xi very likely recognizes that internal investigations can also promote paralysis and degrade organizational cohesion, but views this as a small price to pay for protecting his position.
However, there does not appear to be a clear political rationale behind the most recent removals that would offset the disruption to China’s national security apparatus. The officials who have been targeted played peripheral roles, at best, in the maintenance of Xi’s and the CCP’s political power. The minister of defense is not in the chain of command, and although the Rocket Force is crucial to the PLA’s military capacity, it is probably the most insular service. It plays little role in internal stability operations. The abrupt removals of Qin and senior military leaders so soon after they were put in office therefore exacts a cost to their organizations and to China’s image with little redeeming political value. This suggests shortcomings in the personnel vetting and monitoring systems — and perhaps Xi’s own judgment — that should have highlighted disqualifying factors before they were elevated to positions of prominence.
Implications for Xi’s political and strategic calculus
The recent personnel tumult suggests a variety of implications for Xi and China. First, the clean sweep of the Rocket Force’s leadership team and the imposition of outsiders to replace them indicate that Xi probably does not anticipate fighting a large-scale conflict soon. If he foresaw an imminent likelihood of war, then he probably would not have uprooted the Rocket Force’s entire leadership, or he at least would have chosen replacements more familiar with the force.
The nature of the removals is also a fresh demonstration that the imperatives of secrecy and compartmentalization in party governance are far more important to Xi than assuaging foreign concerns about the CCP’s opaque and seemingly capricious decisionmaking. The silence accompanying the mysterious disappearances of Qin and Li will fuel rumors about other potential targets. It substantiates concerns among foreign officials that their interlocutors are “nowhere near within a hundred miles” of Xi’s inner circle, as US Coordinator for the Indo-Pacific Kurt Campbell described in 2021.
Finally, the extent of the reshuffle will also be a marker of Xi’s broader approach to political control over the elite as he enters a new phase of his tenure. The ebb and flow of investigations over the past eleven years has tracked Xi’s overall political position and priorities. In his first term as general secretary from 2012 to 2017, Xi was consumed by his campaign to target the personal and institutional power bases of his rivals while elevating his allies. After Xi completed his consolidation of power in 2017, the broad purges and reorganizations slowed significantly as he shifted from disruption to construction. Since then, he has focused on rationalizing and strengthening the party apparatus.
The investigations that have come to light recently in Xi’s third term probably reveal cracks in Xi’s personnel management system. However, they do not yet imply a departure from his overall approach to controlling the CCP apparatus or a threat to his power. Xi maintains personal control over the key organs of political power within the party. He relies on a very small circle of trusted subordinates to run those organs and on regular but contained internal investigations into senior officials outside that circle. It would, however, signal a more disruptive and unpredictable approach to governance if leaders and officials at the heart of Xi’s political control apparatus were targeted. Those key organs include the Central Committee’s General Office and Organization Department and the CMC’s General Office. For now, the churn within the upper ranks of the PLA and MFA reflects a familiar, if flawed, playbook for enforcing discipline within the CCP.
Purge mood music: